force building torchvision with CUDA support if CUDA is included as dependency by setting $FORCE_CUDA

[boegel]

(created using eb --new-pr)

If torchvision was built without CUDA support, then the included sanity check command will fail with:

NotImplementedError: Could not run 'torchvision::nms' with arguments from the 'CUDA' backend

See also https://discuss.pytorch.org/t/notimplementederror-could-not-run-torchvision-nms-with-arguments-from-the-cuda-backend-this-could-be-because-the-operator-doesnt-exist-for-this-backend/132352

Setting $FORCE_CUDA bypasses the checks that are done to determine whether or not torchvision should be built with CUDA support.

[boegel]

Test report by @boegel

Overview of tested easyconfigs (in order)

• SUCCESS torchvision-0.13.1-foss-2022a.eb

Build succeeded for 1 out of 1 (1 easyconfigs in total)
node3133.skitty.os - Linux RHEL 8.6, x86_64, Intel(R) Xeon(R) Gold 6140 CPU @ 2.30GHz (skylake_avx512), Python 3.6.8
See https://gist.github.com/boegel/bef4a5353ae4659fdb03ec03457c5a2e for a full test report.

[boegel]

Test report by @boegel

Overview of tested easyconfigs (in order)

• SUCCESS torchvision-0.12.0-foss-2021a-PyTorch-1.11.0-CUDA-11.3.1.eb
• SUCCESS torchvision-0.13.1-foss-2022a-CUDA-11.7.0.eb

Build succeeded for 2 out of 2 (2 easyconfigs in total)
node3307.joltik.os - Linux RHEL 8.6, x86_64, Intel(R) Xeon(R) Gold 6242 CPU @ 2.80GHz (cascadelake), 1 x NVIDIA Tesla V100-SXM2-32GB, 530.30.02, Python 3.6.8
See https://gist.github.com/boegel/44524441c2faccc9f67dfd9f0ceb9b54 for a full test report.

[akesandgren]

LGTM

[akesandgren]

Going in, thanks @boegel!